light770/qwen3-reranker-0.6b

Compact Powerhouse for Vector Reranking

Public
940 runs
Weights
Playground API Examples README Versions

Run time and cost

This model costs approximately $0.00041 to run on Replicate, or 2439 runs per $1, but this varies depending on your inputs. It is also open source and you can run it on your own computer with Docker.

This model runs on Nvidia T4 GPU hardware. Predictions typically complete within 2 seconds.

Readme

Qwen3-Reranker-0.6B is a lightweight yet powerful cross-encoder reranking model from Alibaba’s Qwen team, designed to boost retrieval accuracy in two-stage RAG pipelines. It scores query-document pairs with high precision, perfect for re-ranking top-k results from vector search.

-Cross-Encoder Architecture: Processes query and document together for superior relevance scoring vs. bi-encoders

-Long Context: 32K token context window handles long documents without truncation

-Instruction-Aware: Task-specific instructions improve ranking accuracy by 1–5%

-Multilingual: Supports 100+ languages including code, enabling cross-lingual reranking

Specification Value
Parameters 0.6B (600M)
Architecture Dense Transformer decoder (Causal LM)
Layers 28
Context Length 32,768 tokens
Scoring Method Yes/No token logits
Languages 100+
License Apache 2.0
Release Date June 2025
Model created